RoadTest: RoadTest Review the TI LMZ36002EVM
Evaluation Type: Evaluation Boards
Did you receive all parts the manufacturer stated would be included in the package?: True
What other parts do you consider comparable to this product?: LMZ14202, LM46002PWP EVM
What were the biggest problems encountered?: errors in the data sheet
So its about time to finally write down my review. I'm sorry for the delay - but delivery took long enough that I started another project in the mean time (apart from renovating my study) which also wanted to be completed...
As usual the box this board came in was way too large. TI uses the same box for all its evaluation modules, which for this EVM is already 4 or 5 times too big. And then Farnell also puts it into a large box - in the end one could have put 10 of these board into the same space easily. I liked the old boxes for the MSP430 launchpads, and they would be totally fine here. But I don't pay for packaging and shipping in this case, so I should not complain too loudly.
To get some impressions, I took a photo of the board together with the E14 T-Shirt and the RPi header strip for comparision:
And to compare it with the LM46002 board I reviewed last year I also took a side-by-side photo:
The LM46002 will also serve as comparison object, since it has the same specifications in terms out input voltage range and current capabilities. In terms of total board space both are about comparable, but they differ in their form factor. Both boards have quite some unused space, but when you look at what area is actually needed, the LMZ36002 wins hands-down. The whole chips is about the same size as the inductor needed for the LM46002, and there are much less passives needed.
So the LMZ36002 feels much smaller (even though this board isn't) and in a real application would use much less space space. That is a nice benefit - you get more watts per square meter (or rather square millimeter...)
So lets have a more detailed look at this board:
Inputs and outputs are equipped with nice terminal blocks, but there are no solder pads available. Everything is nicely labeled (which is not true for the LM46002 board - none of the pads on the edge connector is labeled). What I especially like are the header connectors for Vin and Vout - they make properly measuring these signals much easier (more on that later). The board also select different output voltages and switching frequencies. This makes testing different scenarios a breeze.
But when I looked at the other jumper settings, the board, its user guide and the LMZ36002 data sheet confused me. According to the data sheet (page 16), the Versa-Comp pin (V-COMP on the board) can either be tied to Vadj or left floating - depending on the output voltage. But on the EVM, there is no 'floating option', and instead Versa-Comp can be connected to the Sense+ pin. Also, the EVM user guide states that this depends on having additional output capacitance (which the data sheet never mentions). One of these two is wrong here, so I left the jumper connected to Vadj and hoped for the best.
Speaking of mistakes: the LMZ36002 has a nice undervoltage-lockout functionality (UVLO), coupled with the 'enable/inhibit' pin. According to the data sheet, its set for a trigger voltage of 2.1V and should be tied to Vin wuth a 100k resistor. But then the it shows a table to select the other resistor for the divider - and this one is clearly wrong. A 2:1 voltage divider (the first entry in the table) with an input voltage of 4.5V doesn't deliver the needed 2.1V for the UVLO input. In this case the EVM user guide got it correct - it uses a 1:1 resistor divider for its 4.2V UVLO. (And I tested it also with other divider rations, and it worked as one would expect - the data sheet is just wrong here).
Speaking of these tests: one can see in the photo that the EVM provides test points for all signals. The connectors are easy to use with a scope probe, and there are always analog ground signals nearby.
Since I was not chosen as one of the testers for the BK8600 electronic load, I had to roll my own test set up. Testing with a static (slowly changing load) is not as interesting as dynamic load changes. So I went ahead and build a small test load generator using a 555 timer, a TC4421 MOSFET driver and a BUZ10 N-FET to switch between two different loads, with a 1.3Hz frequency):
(The small board at the top is a LT4320 active rectifier - I did not want to use my linear power supply in front of the converter, but instead drive it directly)
To test noise figures and the signal waveforms I used a resistive load (I used the power resistors I found in my parts bin to create a big 4 ohm resistor):
Here you can also see how to use the provider pin header to measure with the scope probe directly (instead of using the hook and the ground clip). See e.g. this article from Texas Instruments.
Last but not least I used the same resistors (and a smaller one) to measure the efficiency of the converter:
(Two UT-61E, one UT-139C and my very first DMM ever - its still quite accurate) Here I used a 20V laptop supply to get a stable input voltage (and because my planned usage involves a 24V input voltage so I wanted to test as close as possible to it). Because of my test setup both current meters are connected in reverse here...
To measure efficiency and compare it with the data sheet values, I measured with 6V, 6V and 2.5V under high load, and with 5V and a small load. For each of these I tested with 300kHz, 500kHz and 1MHz switching frequency. I measured input and output voltage and current, and from there calculated power and efficiency:
With higher output voltages (which result in higher current due to my setup) the efficiency is quite good - up to 90%. The chip got warm but not hot (in contrast to my load resistors...). With the small load its still up to 80% which is not too bad. These figures also match whats shown in the data sheet, so other values can be taken from there.
But one can see two trends here: the efficiency is better with lower switching frequency - the difference between 300kHz and 1MHz is about 5% in all scenarios. This leaves the question why this chip supports higher frequencies? Maybe then it has less ripple? We will see.
The other trend is that the output voltage drops with higher load. I measured the output voltage using the dedicated sense line (at the top right of the board) which is connected directly to the output at the chip, so it should stay constant (apart from the voltage drop on the ground plane). On a 5V output and a 1MHz switching frequency, the difference between the 1.5A and the 35mA load is about 70mV - more than 1% (its less with the other frequencies, so its not only due to the ground currents). This is clearly something to look out for.
But apart from that the LMZ36002 doesn't disappoint. Its clearly able to drive larger loads with high efficiency.
Noise measurements are quite difficult. Its easy to pick up noise from the environment, especially when using a 10x probe as I'm doing (I did not come around to build a simple 1x probe from a RG58 cable). So as a reference I measured the noise floor without anything connected to the probe: about 2-3 mV RMS and 13-16 mV peak-to-peak. (I did the measurements while connecting the board to my linear power supply so I could see the noise introduced by the LMZ36002 itself)
The results are quite encouraging (at least as far as I could measure): noise and ripple are not really measurable. Mostly I measured with 5 V output, using different loads (4 ohm and 140 ohm) and the different switching frequencies. I measured a noise / ripple of about 2.5 up to 5 mV RMS and a peak-to-peak voltage of 14 up to 25 mV. Here one could see the benefit of higher switching frequencies: noise and ripple were significantly lower.
As a comparision, here are the scope screenshots for a 4 ohm load, with 300 kHz and 1MHz switching frequency:
(The scope was set to a bandwith-limit of 20 MHz, so the measured 57.69 MHz are a little bit strange. But the 300 kHz ripple can be seen easily).
The FFT was also showing only the switching frequency as a peak, apart from the the spectrum is quite flat.
Additionally I measured the ripple when connecting the LMZ36002 directly to my bridge rectifier (together with a filter cap of course):
(The blue line is the input voltage showing the 100 Hz ripple of 640 mV). The output ripple can bee seen, but its still not really measurable. I'm quite satisfied with that!
The last test was about dynamic regulation when the load changes, and to look at the behavior during startup. So lets start with the power-on scenario.
Since the LMZ36002 comes with an integrated slow-start functionality, there should be no overshoot when power is applied. To test this I used my active rectifier, and connected it to the transformer (while setting the scope to trigger at the rising flank of the output voltage). The result was this (using 5V output with my 4 ohm load):
The blue line is the input voltage (with 2V/div) and the yellow line is the output (with 1V/div - thats why the input voltage looks lower than the output...). One can see the ripple of the input since its starting directly into a load. The rise time of the output is 8 ms, and there is no overshoot visible. One can also see (measured by the cursors) that the LMZ36002 only starts up when the input voltage rises above 4.5 V - the UVLO uses either the value configured for UVLO (which is 4V) or 90% of the output voltage, whichever is higher. To verify that I configured the UVLO to trigger at 6.5V, and got the expected result.
If you look closely, there is a small bump in the output right at the beginning of the startup. When we zoom in, this is what we get:
Its just 100 mV - probably something is not fast enough to start up, or there is a capacitor that needs to be charged up for proper function. But it doesn't look critical to me.
The LMZ36002 also has a "power good" output signal to notify connected logic that the output is now proper regulated and stable. Lets look at this:
(blue is now the PGOOD output) Yes, looks as it should. It just is a little bit higher than 5V, this is something to look for.
I also compared between running with 300kHz and 1MHz, but found no visible differences.
Lets now look at the behavior during load changes. There my "dynamic load generator" (which is my glorified name for a 555 with a N-FET attached) comes into play. I used it to switch the load from 140 ohm to 4 ohm and back. The scope triggered on the gate of the N-FET, and I probed the output voltage directly at the chip (using the pin headers designated as probe points).
The first screen shot is turning of the load, and the second one for turning on. Note that the output voltage (the blue line) is just 500mV/div, I just changed the offset that much.
So the overshoot when turning of the load is about 200 mV, but the drop when the load is turned on is nearly 400 mV. One can also see that there is a noticeable drop in the output voltage, which looks like about 50 mV. Since I was probing directly at the output the traces on the board should have no effect here.
But the regulation is quite fast - the dropout with an increasing load is regulated after 100 µs, and during turnoff it takes about 200 µs (the output capacitance just needs some time to discharge into the load).
I compared different voltages, different switching frequencies, and also looked at the LM46002 again. So the frequency had no visible effect. But with higher output voltages, the drop with an increasing load got more pronounces. When I set the output to 7.5V, it was close to 600 mV - nearly 10%. This is quite a lot, so probably one needs more output capacitance then.
Speaking of the latter: the LMZ36002 requires at least 64µF of output capacitance, which is quite a lot. Most other DC/DC converters can work with 10µF or so. and just need more to reduce the ripple. The maximum capacitance is limited to 350µF (or less, depending on the voltage), so you need to be careful with distributing additional caps throughout your application.
The LMT46002 I'm using for comparison shows less pronounced over- and undershoot (about 100mV), and also regulates in about half the time. So it seems there is a price to pay for getting such a device in such a small form factor.
One feature I like on the LMZ36002 is its dedicated sense line. Its intended to be connected directly at the load (although the EVM doesn't seem to use it that way) to ensure good regulation. But it can also be used to add an additional current sense resistor between the converter and the load, without affecting the regulation (as it would do otherwise). I think this is quite neat.
And its too bad that the EVM doesn't make proper use of it. It connects the sense line to Vout right between the chip, the output caps and the probe points. And I'm still not convinced that this explains the voltage drops I'm seeing under load.
The slow-start function can also be configured, but it cannot be faster that 4 ms rise time (since then the current limiting would be activated while charging up the output capacitors, and it would shut down).
Well, whats to say about this converter? Surely its delivering quite a high power density (15W in a package smaller than the inductor on the LM46002 EVM board!). But it seems there is a price to pay in term of performance - the LM46002 delivers better and faster load regulation.
I also don't like that the data sheet is contradicting with the EVM user guide. And it has at least the error with the UVLO voltage in it - this raises fears that there are other error as well. (Apart from that its well written, detailed and gives nice guidance how to use it).
I'm surely looking forward to use the EVM board to power the RPi for my 3D printer (which has just a 24V power supply). And I'm sure it will perform well in this application, and the small form factor surely is a plus. If you have something more demanding (esp. with highly dynamic loads) the LM46002 with its similar specs might be a better choice.